nyu-dice-lab/wavepulse-radio-raw-transcripts

WavePulse Radio Raw Transcripts

Dataset Summary

WavePulse Radio Raw Transcripts is a large-scale dataset containing segment-level transcripts from 396 radio stations across the United States, collected between June 26, 2024, and Dec 29th, 2024. The dataset comprises >250 million text segments derived from 750,000+ hours of radio broadcasts, primarily covering news, talk shows, and political discussions.

The summarized version of these transcripts is available here. For more info, visit https://wave-pulse.io

Dataset Details

Dataset Sources

• Source: Live radio streams from 396 stations across all 50 US states and DC
• Time Period: June 26, 2024 - December 29th, 2024
• Collection Method: Automated recording and processing using the WavePulse system
• Audio Processing: WhisperX for transcription and speaker diarization
• Format: Parquet files organized by state and month, with segment-level granularity

Find recordings samples here.

Data Collection Process

1. Recording: Continuous recording of radio livestreams
2. Transcription: Audio processed using WhisperX for accurate transcription
3. Diarization: Speaker separation and identification
4. Quality Control: Automated checks for content quality and completeness
5. Removal of personal information only for cleaning purpose. Radio is fair use.

Dataset Statistics

• Number of Stations: 396
• Number of States: 50 + DC
• Total individual 30-minute transcripts - 1,555,032
• Average Segments per 30-min: ~150
• Total Segments: > 250 million
• Total Words: >5 billion

Usage

Loading the Dataset

from datasets import load_dataset

# Load full dataset
dataset = load_dataset("nyu-dice-lab/wavepulse-radio-raw-transcripts")

# Load specific state
dataset = load_dataset("nyu-dice-lab/wavepulse-radio-raw-transcripts", "NY")

# Filter by date range
filtered_ds = dataset.filter(
    lambda x: "2024-08-01" <= x['datetime'] <= "2024-08-31"
)

# Filter by station
station_ds = dataset.filter(lambda x: x['station'] == 'WXYZ')

# Get all segments from a specific transcript
transcript_ds = dataset.filter(lambda x: x['transcript_id'] == 'AK_KAGV_2024_08_25_13_00')

Data Schema

{
    'transcript_id': str,  # e.g., 'AK_KAGV_2024_08_25_13_00'
    'segment_index': int,  # Position in original transcript
    'start_time': float,   # Start time in seconds
    'end_time': float,     # End time in seconds
    'text': str,          # Segment text
    'speaker': str,       # Speaker ID (unique *within* transcript)
    'station': str,       # Radio station callsign
    'datetime': datetime, # Timestamp in ET
    'state': str         # Two-letter state code
}

Example Entry

{
    'transcript_id': 'AK_KAGV_2024_08_25_13_00',
    'segment_index': 0,
    'start_time': 0.169,
    'end_time': 2.351,
    'text': 'FM 91.9, the Nana.',
    'speaker': 'SPEAKER_01',
    'station': 'KAGV',
    'datetime': '2024-08-25 13:00:00',
    'state': 'AK'
}

Important Notes

• Speaker IDs (e.g., SPEAKER_01) are only unique within a single transcript. The same ID in different transcripts may refer to different speakers.
• Segments maintain their original order through the segment_index field.
• All timestamps are relative to the start of their 30-minute transcript.

Data Quality

• Word Error Rate (WER) for transcription: 8.4% ± 4.6%
• Complete coverage of broadcast hours from 5:00 AM to 3:00 AM ET (i.e. 12 AM PT)
• Consistent metadata across all entries
• Preserved temporal relationships between segments

Intended Uses

This dataset is designed to support research in:

• Media analysis and content tracking
• Information dissemination patterns
• Regional news coverage differences
• Political narrative analysis
• Public discourse studies
• Temporal news analysis
• Speaker diarization analysis
• Conversational analysis
• Turn-taking patterns in radio shows

Limitations

• Limited to stations with internet streams
• English-language content only
• Coverage varies by region and time zone
• Potential transcription errors in noisy segments
• Some stations have gaps due to technical issues
• Speaker IDs don't persist across transcripts
• Background music or effects may affect transcription quality

Ethical Considerations

• Contains only publicly broadcast content
• Commercial use may require additional licensing
• Attribution should be given to original broadcasters
• Content should be used responsibly and in context

Citation

@article{mittal2024wavepulse,
  title={WavePulse: Real-time Content Analytics of Radio Livestreams},
  author={Mittal, Govind and Gupta, Sarthak and Wagle, Shruti and Chopra, Chirag and DeMattee, Anthony J and Memon, Nasir and Ahamad, Mustaque and Hegde, Chinmay},
  journal={arXiv preprint arXiv:2412.17998},
  year={2024}
}